River Runoff Influences on the Central Mediterranean Overturning Circulation

Clim Dyn DOI 10.1007/s00382-017-3715-9

River runoff influences on the Central Mediterranean overturning circulation

Giorgia Verri1 · N. Pinardi1,2 · P. Oddo3,4 · S. A. Ciliberti1 · G. Coppini1

Received: 15 June 2016 / Accepted: 28 April 2017 © The Author(s) 2017. This article is an open access publication

Abstract The role of riverine freshwater inflow on the dominant forcing mechanism and the potential role of river Central Mediterranean Overturning Circulation (CMOC) runoff has to be considered jointly with wind work and heat was studied using a high-resolution ocean model with a flux, as they largely contribute to the energy budget of the complete distribution of rivers in the Adriatic and Ionian basin. Looking at the downwelling branch of the CMOC catchment areas. The impact of river runoff on the Adri- in the Adriatic basin, rivers are demonstrated to locally atic and Ionian Sea basins was assessed by a twin experi- reduce the volume of Adriatic dense water formed in the ment, with and without runoff, from 1999 to 2012. This Southern Adriatic Sea as a result of increased water strati- study tries to show the connection between the Adriatic fication. The spreading of the Adriatic dense water into the as a marginal sea containing the downwelling branch of Ionian abyss is affected as well: dense waters overflowing the anti-estuarine CMOC and the large runoff occurring the Otranto Strait are less dense in a realistic runoff regime, there. It is found that the multiannual CMOC is a persis- with respect to no runoff experiment, and confined to a nar- tent anti-estuarine structure with secondary estuarine cells rower band against the Italian shelf with less lateral spread- that strengthen in years of large realistic river runoff. The ing toward the Ionian Sea center. CMOC is demonstrated to be controlled by wind forcing at least as much as by buoyancy fluxes. It is found that river runoff affects the CMOC strength, enhancing the amplitude 1 Introduction of the secondary estuarine cells and reducing the intensity of the dominant anti-estuarine cell. A large river runoff can The annual mean freshwater budget in the Mediterranean produce a positive buoyancy flux without switching off the Sea, composed of evaporation minus precipitation and river antiestuarine CMOC cell, but a particularly low heat flux runoff, has been found to be positive, corresponding to a and wind work with normal river runoff can reverse it. net surface loss of 0.54 ± 0.15 m/year as the average of Overall by comparing experiments with, without and with several long term annual estimates proposed by Sanchez- unrealistically augmented runoff we demonstrate that rivers Gomez et al. (2011). Moreover, the net heat budget of the affect the CMOC strength but they can never represent its basin is well known to be negative according to most estimates (Bethoux 1979; Pettenuzzo et al. 2010; Sanchez- Gomez et al. 2011; Schroeder et al. 2012). The combina- * Giorgia Verri tion of the positive freshwater budget and the negative heat [email protected] budget of the Mediterranean Sea gives a net buoyancy loss, 1 Centro EuroMediterraneo sui Cambiamenti Climatici, Via A. thus sustaining a vigorous mean kinetic energy in the basin Imperatore 16, 73100 Lecce, Italy (Cessi et al. 2014). 2 Universita’ degli Studi di Bologna, Dipartimento di Fisica e This forcing induces also a vigorous meridional anties- Astronomia, Bologna, Italy tuarine overturning circulation. Pickard and Emery (1982) 3 Istituto Nazionale di Geofisica e Vulcanologia, Bologna, Italy differentiate between the semienclosed seas with horizon- 4 Present Address: Center for Maritime Research tally and vertically separated inflow and outflow regimes and Experimentation CMRE, La Spezia, Italy at the strait. Semienclosed basins with strait flows that

Vol.:(0123456789)1 3 G. Verri et al. are vertically separated can have two vertical circulation large LIW inflow from the Levantine basin. This Aegean modes, estuarine and antiestuarine. Meridional Overturning Circulation (AeMOC) character- Traditionally, the estuarine and anti-estuarine circula- izes the Eastern Mediterranean Transient (EMT) which tion in semienclosed seas has been classified on the basis of started at the end of the eighties and ended around the the surface water flux and the conservation of salt (Knud- middle of the nineties. Gacic et al. (2011) pointed out sen 1900; Sverdrup 1947; Pickard and Emery 1982). More that the EMT could be a repeating phenomenon, and part precisely, estuarine and anti-estuarine circulations can then of its cause is connected with inversions of the Ionian be represented by meridional overturning cells which form circulation. into the semi-enclosed basin and which are connected to The CMOC occupies the Northern Ionian Sea and the external regions by the Strait inflow/outflow system. extends into the Southern Adriatic (Fig. 1). The lower The general characteristics of the Mediterranean over- branch of the CMOC is connected with the outflowing of turning circulation are schematized in Fig. 1 (Pinardi the Southern Adriatic dense waters (Artegiani et al. 1997a; et al. 2006). This circulation is characterized by inter- Vilibić and Orlić 2001–2002; Manca et al. 2002; Mantzi- annual as well as multi-decadal time scales and it is afou and Lascaratos 2004, 2008; Wang et al. 2007; Bensi composed of three major conveyor belts: the Zonal Over- et al. 2013b) in the Ionian abyssal basin. Several authors turning Circulation (ZOC) in the Southern Mediterra- discussed that, during the EMT, the Adriatic source of Ion- nean propelled by the Gibraltar stream flow and Levan- ian abyssal waters could have been replaced by the Cre- tine Intermediate Water (LIW) formation processes, the tan Sea deep waters (Roether et al. 1996; Roussenov et al. Western Mediterranean Meridional Overturning Cir- 2001; Curchitser et al. 2001; Manca et al. 2003; Rubino and culation (WMOC) originating in the Gulf of Lion, the Hainbucher 2007; Ursella et al. 2011; Rubino et al. 2012; Central Mediterranean MOC originating in the Adriatic Bensi et al. 2013a). Sea (CMOC). These overturning cells are triggered by In this paper we would like to exhamine for the first time buoyancy losses and water mass sinking occurring in the the question of what is the influence of river runoff on the regions offshore the Gulf of Lions, in the Southern Adri- CMOC. The question is relevant because more than 30% atic and in the Northern Levantine Basin, enhanced by the of the Mediterranean runoff is concentrated in the Adriatic presence of large scale permanent cyclonic gyres driven Sea and the hypothesis that large freshwater runoff in the by wind stress curl (Pinardi et al. 2006). The Aegean Sea Adriatic could shutdown the CMOC is realistic at least for is marked as another dense water site in Fig. 1: Roether the past. As far as we know this is the first study on the et al. (1996) and Gertman et al. (2006) demonstrated that river influence on the CMOC, similar to Rahmstorf’s study dense waters could form there during intense winters and (Rahmstorf 1995) on the freshwater role in the Northern Atlantic overturning circulation. Previously, Skliris et al. (2007) and Vervatis et al. (2013) investigated the impact of reduced discharge of Ebro and Nile rivers on the dense water formation in the Eastern and Western Mediterranean sub-basins. More generally, the role of freshwater inputs (due to both rivers and precipitations) on the dynamics of a marginal sea, such as the Adriatic basin, has been widely investigated in the literature. Spall (2012) demonstrates that, in marginal sea areas, an increase in surface freshwater gain can lead to a shutdown of dense water formation and sinking, and the marginal sea MOC switches from anti- estuarine to estuarine mode. Recently, Cessi et al. (2014) established that the estuarine/anti-estuarine character of Fig. 1 The conveyor belts of the Mediterranean Sea. The red and a semi-enclosed sea with a two-layer flow at the strait is yellow dashed streamlines in the zonal direction stand for the zonal determined by both wind and buoyancy forcings. The wind overturning circulation in the surface-intermediate layers that is forcing is normally a source of mechanical energy for the forced by the Gibraltar stream flow and Levantine Intermediate Water (LIW) formation processes. The red spirals indicate the preferential circulation, while the buoyancy forcing could be either an sites for strong heat losses during wintertime and dense water forma- energy source or a sink depending on the sign of the net tion processes. Two anti-cyclonic meridional overturning circulation buoyancy flux at the surface. For estuarine basins, such as white spirals patterns can be distinguished ( ): the Western Mediter- the Baltic and Black Sea, the positive buoyancy flux (domi- ranean MOC originating in the Gulf of Lion, and the Central Medi- terranean MOC originating in the Adriatic Sea. Reproduced from nated by precipitation and runoff exceeding evaporation), is Pinardi et al. (2006) a net energy sink for the circulation, thus producing a less

1 3 River runoff influences on the Central Mediterranean overturning circulation vigorous meridional circulation than in the anti-estuarine Sea is conventionally subdivided on the basis of its bottom basins. morphology: the Northern (NAd), the Middle (MAd) and The Adriatic Sea is a “dilution basin” with a nega- the Southern Adriatic (SAd). The connection with the Ion- tive annual freshwater budget, equal to the difference ian Sea occurs at the Otranto Strait where the sill is 800 m between evaporation and precipitation and runoff, of deep, located at approximately 40°N. about −1 m year−1 (Artegiani et al. 1997a, b) but a nega- This model is the first implementation of the NEMO tive annual mean heat budget. The buoyancy flux could code over the Central Mediterranean Sea area, with a rep- eventually then be positive, thus determining a net sink of resentation of almost all the rivers flowing into the Cen- energy and possibly a net estuarine character of the circu- tral Mediterranean Sea. In this work the regional model lation. Pinardi et al. (2006) show that, due to river runoff, is forced by a 1/16° resolution daily analyses from the the Adriatic Sea could be characterized by zero net buoy- operational Mediterranean forecasting system, MFS ancy flux, thus producing a basin where the circulation is (Tonani et al. 2008; Pinardi and Coppini 2010) at the lat- mainly powered by the wind stress. However, the energet- eral open boundaries. The numerical model configuration ics proposed by Cessi et al. (2014) cannot be applied sat- is explained in details in “Appendix 1”. isfactorily to the Adriatic Sea, since the flow at Otranto is A set of experiments were performed with present day not just a two-layer flow. Thus, a comprehensive analysis of river runoff, without river runoff and with augmented run- surface buoyancy and the CMOC is needed to fully estab- off (Table 1), spanning the period from 1 January 1999 to lish the estuarine/anti-estuarine character of the Adriatic 31 December 2012. The time series of the kinetic energy Sea circulation. integrated over the basin volume (not shown) shows that In order to answer the effects of runoff on the CMOC the spin up period consists of the first few months of 1999, and the Adriatic Sea deep water formation mechanisms, thus this year has been eliminated from the analysis. Such a high resolution general circulation model was set up, short spin up is due to the fact that the model was initial- forced by realistic fluxes of water, heat and momentum. ized by close-to-present-day fields from MFS model. The impact of river runoff on the circulation is assessed by It is worth to stress that the conceptual paradigm of this using a mechanistic approach: a full forcing and dynamics study is to explain the “theoretical role” of river inflow experiment is compared to a zero-runoff case. The full forc- since the case without or with increased runoff corresponds ing experiment is considered to be the present state of the to extreme conditions with respect to the present day. EXP1 Adriatic Sea and it is validated with existing data sets. corresponds as closely as possible to reality and this is The paper is organized as follows. Section 2 outlines the our reference against which we analyze the effects only of experimental design, and details the parameterization of river runoff changes, leaving all the other forcings identi- the rivers as surface boundary conditions and the validation cal because we know that the estuarine water balance of the of the model performance. Section 3 describes the experi- Adriatic is due to river runoff. ments highlighting the role of the river runoff forcing. A summary and conclusions are presented in the last section. 2.2 River runoff datasets

River runoff into the Mediterranean Sea is mainly concen- 2 Experimental design trated in the Central Mediterranean sub-basin, with rivers flowing into the Adriatic Sea providing almost 1/3 of 2.1 Model configuration and experiments set‑up the total runoff (Struglia et al. 2004; Ludwig et al. 2009). Overall the Mediterranean Sea counts on a great num- The numerical model used is the Nucleus for European ber of very small rivers (Milliman 2001), owing to strong Modelling of the Ocean, NEMO (Madec 2008), that is a topographic relief favouring the formation of small water- three-dimensional finite difference numerical model adopt- sheds. The construction of dams in the Nile and Ebro have ing the Boussinesq and hydrostatic approximations and reduced their runoff in the Mediterranean (Nile from 2700 using the linear explicit free surface formulation. The area to 150 m3 s−1, Ebro from 1500 to 400 m3 s−1) so that cur- covered by the model grid is the Central Mediterranean Sea rently the only major runoff sources located out of the Cen- from 12.2 to 21.0°E and 30.2 to 45.8°N with a horizon- tral Mediterranean Sea are the Rhone and the Ebro rivers tal resolution of about 2.2 km (2.5 km in the meridional flowing into the Western sub-basin. direction and 1.7–2.2 km in the zonal direction). Figure 2 The freshwater discharge into the Central Mediterra- shows the bathymetry of the domain, river mouth grid nean Sea is almost totally concentrated along the Adriatic points and the three sub-regions into which the Adriatic coastlines: the Po river is the main freshwater source of the

1 3 G. Verri et al.

Fig. 2 Model domain and details on areas of interest. The red lines define the three Adriatic sub-regions and the Ionian Sea. Black isolines show the bathymetry. Blue stars and arrows indicate the model river mouths

1 3 River runoff influences on the Central Mediterranean overturning circulation

Table 1 List of experiments Experiment River discharge and the runoff values assumed EXP 1 Realistic river discharge: 67 surface sources of climatological runoff (except daily observations for the Po river runoff) and constant salinity EXP 2 No river discharge EXP 3 Mean annual discharge augmented of 50% with respect to Exp1 EXP 4 Mean annual discharge augmented of 100% with respect to Exp1

Central Mediterranean Sea and accounts for almost 30% climatologies and the annual mean discharges as useful ref- of the Adriatic annual discharge (Cushman-Roisin et al. erence values. 2002). Besides the Po, other significant freshwater inputs All river mouths are “point sources” except for 2: are the Buna/Bojana, Vjose and Neretva along the Eastern Marecchia to Tronto rivers (Tronto excluded) in the Adriatic coast, and the Adige and Isonzo along the northern Marche region and Vibrata to Fortore rivers (Fortore Italian coast. excluded) in the Abruzzo and Molise regions which are Several Mediterranean river discharge datasets have “diffused sources”, i.e. the runoff was split across sev- started to be collected during the last decade. Ludwig eral grid points. These diffused sources and the rivers et al. (2009) proposed a reconstruction of the river runoff in Puglia are the only rivers of the model based on Rai- of the Mediterranean and Black Seas during the period cich’s (1996) climatologies (see Fig. 2 for locating river 1960–2000, on the basis of available observations and outlets). empirical modeling for rivers and time periods where The Po river runoff, referred to the Pontelagoscuro observations were missing. Ludwig et al. (2009) time series station, usptream of the delta, is unequally subdivided considers 295 Mediterranean and Black Sea rivers with the between the nine grid points representing the nine branches condition they cover at least 20 years as reference period. of the delta (Po di Goro, Po di Gnocca, Po di Tolle, Po di Many of the records were downloaded by Mediterranean Bastimento, Po di Scirocco, Po di Bonifazi, Po di Dritta, Po Hydrological Cycle Observing System, Medhycos, data di Tramontana, Po di Maistra) according to percentages in server (Medhycos 2001), other from the Global River Dis- Provini et al. (1992). charge database RivDis (Vörösmarty et al. 1996), or from Daily time series of total discharge in the model domain the Hydro database at the French Ministry of Environment during the simulation period (1 January 1999–31 Decem- (Hydro 2006). Struglia et al. (2004) provides an extensive ber 2012), are shown in the top panel of Fig. 3, with over- representation of historical river release into the Mediter- lapped Po daily discharge as well as the contribution of all ranean basin, timeseries of 67 rivers are derived from the the other rivers (i.e. monthly climatologies interpolated on Global Runoff Data Center (GRDC) hydrological database daily basis). Middle panel highlights the added value of and the Medhycos database. working with daily observations of Po river instead of con- Most of the datasets we adopted consist of observa- sidering monthly climatologies. Finally the bottom panel tions taken at the hydrometric stations nearest to river shows how the Adriatic river release, Po river excluded, is mouths, and few of them are estimated values (Pasaric represented by Raicich’s climatologies (1996), Ludwig’s 2004; Malacic and Petelin 2009; Raicich 1996). Time climatologies (Ludwig et al. 2009) and the new climatolo- series for the various rivers cover different periods. How- gies we chose in this study on the basis of more updated ever, the time series of the major rivers, accounting for the and reliable datasets. To note that our dataset includes 52 most of the Central Mediterranean discharge, overlap at rivers flowing into the Adriatic Sea, while Raicich (1996) least for 20 years. The databases we considered have been considers only 45 and Ludwig et al. (2009) only 32 Adri- provided by the GRDC, the regional agency Autorita’ di atic rivers. Bacino Basilicata, the hydrological division of the National Adding all the contributions, the annual average Research Council CNR IRPI, the regional agencies for the runoff in the Central Mediterranean Sea is equal to environmental protection ARPAs, the Albanian Hydrome- 4.72 × 103 m3 s−1, 29.7% coming from the Po river only, teorological Institute. 94.6% is the contribution of all the Adriatic rivers and 5.4% Overall this study considers 67 Adriatic and Ionian riv- the one of the Ionian rivers. Maximum values of the total ers in total, 52 flowing into the Adriatic Sea and 15 into the discharge on daily basis were observed in autumn 2000, Ionian Sea. Table 2 lists the adopted climatological datasets autumn 2002, and autumn 2008–winter 2009. Annual for river runoff, the time range for computing the monthly mean peaks took place in 2000 (5.28 × 103 m3 s−1), 2002

1 3 G. Verri et al.

Table 2 River runoff climatological values adopted for the Adriatic and Ionian rivers, time period for the climatologies and mean annual discharge values River Dataset Reference period Discharge basin Mean annual discharge (m3 s−1)

Adige (Italy) GRDC 1922–1984 Adriatic Sea 223.75 Agri (Italy) Autorita’ di bacino Basilicata n.r Ionian Sea 9.14 Alcantara (Italy) Piano Tutela Acque Sicilia 1980–1997 Ionian Sea 4.7 Arachthos (Greece) GRDC 1964–1980 Ionian Sea 19.73 Basento (Italy) CNR IRPI 1933–1971 Ionian Sea 13.23 Belice (Italy) Piano Tutela Acque Sicilia 1980–2000 Ionian Sea 51.57 Bevano (Italy) ARPA EMR n.r Adriatic Sea 6 Bistrica (Albania) Albanian Hydrometeorological Institute 1949–1987 Ionian Sea 32.1 Bocca di Primero (Italy) ARPA VENETO n.r Adriatic Sea 10.28 Bradano (Italy) CNR IRPI 1929–1971 Ionian Sea 5.85 Brenta (Italy) ARPA VENETO n.r Adriatic Sea 93.17 Buna/Bojana (Albania-Montenegro) Albanian Hydrometeorological Institute 1965–1985 Adriatic Sea 675 Canale dei Lovi (Italy) ARPA VENETO n.r Adriatic Sea 22.7 Canale di Morgo (Italy) ARPA VENETO n.r Adriatic Sea 10.28 Canale Nicessolo (Italy) ARPA VENETO n.r Adriatic Sea 22.7 Cervaro (Italy) Raicich (1996) n.r Adriatic Sea 2.92 Cetina (Croatia) Pasaric (2004) 1947–2000 Adriatic Sea 88.28 Crati (Italy) CNR IRPI 1926–1966 Ionian Sea 26.2 Dubracina (Croatia) Pasaric (2004) 1947–2000 Adriatic Sea 4.14 Erzen (Albania) Albanian Hydrometeorological Institute 1949–1992 Adriatic Sea 16.9 Fiumi Uniti (Italy) ARPA EMR n.r Adriatic Sea 12.06 Fortore (Italy) Raicich (1996) n.r Adriatic Sea 12.25 Imera Meridionale (Italy) GRDC 1978–1980 Ionian Sea 4.26 Ishm (Albania) Albanian Hydrometeorological Institute 1968–1992 Adriatic Sea 19.8 Isonzo (Italy) Malacic and Petelin (2009) 1945–2000 Adriatic Sea 110.43 Jadro (Croatia) Pasaric (2004) 1947–2000 Adriatic Sea 7.18 Krka (Croatia) Pasaric (2004) 1947–2000 Adriatic Sea 56.51 La Fosa (Italy) ARPA VENETO n.r Adriatic Sea 10.28 Lamone (Italy) ARPA EMR n.r Adriatic Sea 12.06 Livenza (Italy) ARPA VENETO n.r Adriatic Sea 88.33 Marecchia to Tronto, Tronto excluded (Italy) Raicich (1996) 1956–1965 Adriatic Sea 121.92 Mat (Albania) Albanian Hydrometeorological Institute 1951–1986 Adriatic Sea 87.4 Mirna (Croatia) Pasaric (2004) 1947–2000 Adriatic Sea 7.91 Neretva (Croatia) Pasaric (2004) 1947–2000 Adriatic Sea 366.86 Neto (Italy) ARPA CAL n.r Ionian Sea 6.22 Ofanto (Italy) Raicich (1996) n.r Adriatic Sea 14.92 Ombla (Croatia) Pasaric (2004) 1947–2000 Adriatic Sea 27 Pavla (Albania) Albanian Hydrometeorological Institute 1951–1991 Ionian Sea 6.69 Piave (Italy) ARPA VENETO n.r Adriatic Sea 54.33 Platani (Italy) GRDC 1978–1980 Ionian Sea 2.37 Po di Levante (Italy) ARPA EMR n.r Adriatic Sea 21.67 Po di Volano (Italy) ARPA EMR n.r Adriatic Sea 6 Pto Buso (Italy) ARPA VENETO n.r Adriatic Sea 10.28 Pto di Chioggia (Italy) ARPA VENETO n.r Adriatic Sea 17.27 Pto di Lido (Italy) ARPA VENETO n.r Adriatic Sea 17.27 Pto di Malamocco (Italy) ARPA VENETO n.r Adriatic Sea 17.27 Pto Lignano (Italy) ARPA VENETO n.r Adriatic Sea 10.28

1 3 River runoff influences on the Central Mediterranean overturning circulation

Table 2 (continued) River Dataset Reference period Discharge basin Mean annual discharge (m3 s−1)

Rasa (Croatia) Pasaric (2004) 1947–2000 Adriatic Sea 1.58 Reno (Italy) ARPA EMR n.r Adriatic Sea 49.33 Rjecina (Croatia) Pasaric (2004) 1947–2000 Adriatic Sea 7.22 Rubicone (Italy) ARPA EMR n.r Adriatic Sea 6 Savio (Italy) ARPA EMR n.r Adriatic Sea 12.06 Seman (Albania) Albanian Hydrometeorological Institute 1948–1987 Adriatic Sea 86 Shkumbi (Albania) Albanian Hydrometeorological Institute 1948–1991 Adriatic Sea 58.7 Sile (Italy) ARPA VENETO n.r Adriatic Sea 52.92 Simeto (Italy) GRDC 1978–1980 Ionian Sea 3.31 Sinni (Italy) CNR IRPI 1937–1976 Ionian Sea 20.58 Tagliamento (Italy) ARPA VENETO n.r Adriatic Sea 96.92 Thyamis (Greece) GRDC 1963–1978 Ionian Sea 51.39 Tronto (Italy) Raicich (1996) 1956–1965 Adriatic Sea 17.92 Uso (Italy) ARPA EMR n.r Adriatic Sea 6 Vibrata to Fortore + Pescara + San- Raicich (1996) 1956–1965 Adriatic Sea 190 gro + Trigno + Biferno (Italy) Vjiose (Albania) Albanian Hydrometeorological Institute 1948–1987 Adriatic Sea 189 Zellina (Italy) ARPA VENETO n.r Adriatic Sea 10.28 Zrmanja (Croatia) Pasaric (2004) 1947–2000 Adriatic Sea 40.10 Zrnovnica (Croatia) Pasaric (2004) 1947–2000 Adriatic Sea 1.76

Some of the datasets consist of observations taken at hydrometric stations and some are estimated values. To note that Po di Levante and Po di Volano are point sources different from the nine branches of the Po delta. Po river runoff values are not included in this Table since daily averaged observations are assumed at each branch of the delta based on Pontelagoscuro station

(5.19 × 103 m3 s−1), 2009 (5.37 × 103 m3 s−1), and 2010 conditions respectively. Here we follow the natural (5.26 × 103 m3 s−1). The daily and annual maximum val- boundary condition approach (Huang 1993) plus ad-hoc ues are due to Po river only, as the other rivers consist of salt values at river outlets, this is the most consistent monthly climatologies. As detailed above, the Po river rep- before rivers will be considered with lateral open bound- resents the largest contribution to the total discharge of the ary conditions. Such boundary condition involves both a Central Mediterranean Sea, however we should consider water and salt flux condition, as written in Eqs. (20) and we are missing the daily and interrannual variability com- (21), in order to conserve the volume integrated salt con- ing from the other rivers. tent in the basin. It can be easly demonstrated that by tak- The interannual variability described above is the result ing the volume integral of advection/diffusion equation of only Po river runoff interannual differences because for salinity and by replacing the boundary conditions (20) daily data sets for other rivers are not available for such an and (21) we obtain the salt conservation regardless of the extended time period. constant value of salinity chosen at river mouths. Kourafalou et al. (1996) implemented for the first time 2.3 River parameterization in the model the natural boundary condition to the riverine freshwater flux, followed by Skliris et al. (2007), Beuvier et al. Model rivers are parameterized as “surface sources” of (2010), Dell’Aquila et al. (2012) and Vervatis et al. water at the estuary border grid points while no tempera- (2013). In addition operational models of the Mediterra- ture information is prescribed. Our assumption of no tem- nean Sea already use the natural boundary condition from perature differences between river inflow and the marine several years (Oddo et al. 2009). environment is generally valid as river plumes are con- Following Simoncelli et al. (2011) we prescribe controlled by the salinity gradients. stant salinity values at all river mouths parametrizing the The runoff and salinity values are prescribed at river effects of tidal mixing inside the river estuaries. Values outlets in the vertical velocity and salt flux boundary chosen are equal to 15 psu for all rivers, except for the Po

1 3 G. Verri et al.

Fig. 3 River discharge into the Central Mediterranean Sea during the entire simulation period, 1999-01-01 to 2012-12-31. Top panel Daily time series of total discharge. Black line is the discharge of all rivers flowing into the model domain. Green line is the discharge of Po river, which is the only one based on ARPA ERM daily observations. Magenta line is the contribution of all rivers, except Po, which are based on monthly climatologies. Middle panel Daily time series of Po river discharge. Comparison among observed runoff and daily interpolation of 10000 Po runoff by ARPA daily obs Raicich’s climatologies (1996) 9000 Po runoff by Raicich climatologies (1996) and Ludwig’s climatologies Po runoff by Ludwig climatologies(2009) (Ludwig et al. 2009). Bottom 8000 panel Climatologies of the ) 7000 /s Adriatic rivers. Comparison 3 of our model dataset with Rai- 6000 cich’s and Ludwig’s ones 5000

4000

Po Discharge (m 3000

2000

1000

0 Jan99Jan00 Jan01Jan02 Jan03Jan04 Jan05Jan06 Jan07 Jan08 Jan09Jan10 Jan11Jan12 Day

6000

5500

5000

4500

4000 ) /s 3 3500

3000

2500 Discharge (m 2000

1500

1000

500 Adriatic runoff climatologies by this paper Adriatic runoff climatologies by Raicich 1996 Adriatic runoff climatologies by Ludwig 2009 0 JanFeb MarApr MayJun JulAug SepOct NovDec Month

river where 17 psu is considered. These constant salin- 2.4 Model validation with observations ity values are the result of sensitivity tests performed on the basis of salinity profiles measured at river mouths The performance of EXP1 was evaluated by comparing the (Simoncelli et al. 2011) and at the center of the basin simulated fields with available in situ observations. For the (Oddo et al. 2005). in situ data, two Argo profiling floats in the Adriatic and Ion- ian Seas (Fig. 4) were used to calculate vertical mean profiles

1 3 River runoff influences on the Central Mediterranean overturning circulation

Argo Float Path 3 Discussion on the results 46oN

3.1 Is the Adriatic Sea an estuarine or anti‑estuarine basin?

44oN The literature contains ample evidence that the MOC is primarily driven by wind and tidal stirring (Munk and 145 8595 135 Wunsch 1998; Paparella and Young 2002; Marshall and 155 105 115185193 125 75 Speer 2012). In addition, the relationship between the 65 16175 5 55 5 1 45 dense water formation, driven by buoyancy fluxes, and 42oN 15 the strength of the overturning circulation has been high- 25 35 lighted in several theoretical, as well as realistic modelling studies (Rahmstorf 1995, 1996; Pisacane et al. 2006). latitude (°N) Similarly, here we focus on the downwelling branch of

40oN the Central Mediterranean MOC, which develops inside 92 45 the Southern Adriatic sub-basin due to the local open- 65 7555 35 ocean convection and dense water formation sustained 85

25 by winter heat losses and a cyclonic gyre driven by wind 51 15 stress curl. o 38 N In this section we will describe both the surface forcing and the newest theoretical framework which connects 12oE 14oE 16oE 18oE 20oE the anti-estuarine and estuarine character of a marginal longitude (°E) sea to the vertical overturning circulation.

Fig. 4 The trajectory of two Argo profiling floats over 2010–2012. Numbers indicate the ascending profiles: 193 into the Adriatic Sea 3.1.1 The surface forcing and 92 into the Ionian Sea In order to assess the Adriatic Sea overturning circulation on the basis of surface forcing, an analysis of both buoy- of RMSE and BIAS for temperature and salinity (Fig. 5). ancy and wind stress fluxes was performed. The surface 2 −3 The profiles are in the area of deep water formation and in buoyancy flux per unit area,Q B (m s ) is calculated the northern Ionian exit, so in the most relevant places for our according to Cessi et al. (2014) as follows: simulations. On the whole, EXP1 is in good agreement with g T the observed data, comparable RMSE and BIAS values were Qb = Q − SS0g(E − P − R∕A) wCw (1) found for different models of the Adriatic Sea (Guarnieri et al. 0 2013) and the Ionian Sea (Federico et al. 2016; Kassis et al. where T,S are the coefficients of thermal and haline expan- 2016). In EXP2 without rivers, water masses in the Adriatic sion respectively, 0w is the reference sea surface water den- Sea appear saltier and warmer with respect to the observed sity, Q is the net heat flux,C w is the heat capacity of sea dataset. EXP2 has then doubled the RMSE and BIAS errors water, S0 is the surface salinity. Finally, (E − P − R∕A). is with respect to observations thus showing already a large the freshwater flux with evaporation rate, E, and precipita- impact of rivers on the estimate of the circulation in the Adri- tion rate P, in m s−1, river discharge R in m3 s−1 and A rep- atic Sea. The mean profiles of RMSE and BIAS do not show resenting the grid area of river mouths (m 2). Furthermore −4 −1 significant differences in the Ionian Sea instead. the following values were assumed: T = 2.3 × 10 C , −4 −1 −1 ◦ −1 Basin-averaged timeseries of analysed and modelled sea S = 7.5 × 10 psu , Cw = 3990 J kg C , −3 surface temperature are shown in Fig. 6. Analysed SST is S0 = 38.7 psu, 0w = 1029 kg m . The net heat flux, Q, obtained by Optimal Interpolation of Pathfinder AVHRR components are computed according to bulk formulae (Advanced Very High Resolution Radiometer) SST data (Pis- described in “Appendix 1”. ano et al. 2016). Model results are characterized by a positive Our results show the the whole Central Mediterranean BIAS but are capable to reproduce the observed interannual Sea has a positive freshwater budget, 0.60 m year−1, while variability. The agreement between model and analyses time- the Adriatic Sea is negative, −0.69 m year−1, as expected. series is high, with correlations of about 99.6% and RMSE is The annual time series of Adriatic surface buoyancy 0.8 °C for EXP1. flux and wind work is shown in Fig. 7. The realistic

1 3 G. Verri et al.

Fig. 5 Temperature and Salinity RMSE and BIAS for the available Argo observations over 2010–2012 in EXP1 and EXP2 buoyancy flux (EXP1) is generally negative, implying a the deep water formation processes in the Southern Adri- net anti-estuarine forcing of the circulation, however in atic Sea. The green line in Fig. 7, top panel shows the certain years, the values can be several times smaller and difference between the experiments, EXP1–EXP2, and even change sign. This is the case for 2000, 2002 and reaches the maximum value in years 2002–2003, this 2008 where the buoyancy flux reaches small or positive means that in 2002 and 2003 the Adriatic buoyancy values. We argue that small or positive buoyancy budget budget is mainly influenced by river runoff with respect could weaken the anti-estuarine CMOC, as well as reduce to the other forcing mechanisms. A strong river release

1 3 River runoff influences on the Central Mediterranean overturning circulation

Fig. 6 Monthly time series of satellite (black line) and modelled (red line for EXP2 and blue line for EXP1) sea surface temperature

Fig. 7 Top panel Annual time series of buoyancy flux, Qb ∬ dA∕A, for EXP1 (blue line) and EXP2 (red line) and the relative difference. Bottom panel Annual time series of normalized1 averaged⋅u dA wind ∬ s work, 0 V , for EXP1 and EXP2. Results are only relevant to the Adriatic Sea

⋅ occurred also in 2009 (see Fig. 3) but this year is charac- 3 −3 Tw us u The wind work (m s ) is defined as 0 where s terised also by a intense wind work (Fig. 7, bottom panel) which makes river influence on the basin boyancy budget is the sea surface velocity, 0 is the reference sea surface T less relevant. water density and w is the wind stress defined in

1 3 G. Verri et al.

“Appendix 1”. The heat, freshwater and wind forcings act- Spall (2012) also shows that the non-dimensional ratio ing at the air–sea interface may be directly compared by ΔS∕ΔT, where ΔT and ΔS are respectively the model-diag- considering the buoyacy flux and the wind work: both nosed temperature and salinity non-dimensional anomalies quantities represent source/sink terms of energy. In order to between the interior basin and the open boundary currents, compare the surface integrated buoyancy flux with the sur- can be written as a function of the thermal and freshwa- face integrated wind work, the latter is integrated over the ter forcing parameters (see “Appendix 2” for details on the basin area and divided by the basin volume. Figure 7 shows computation of these values). A ratio ΔS∕ΔT less than 1 that the wind work is always positive (10−8 m2 s−3), imply- means that the general circulation is in “thermal mode”, ing a net of mechanical energy for the Adriatic Sea that is which means that the heat forcing prevails and an anti- one order of magnitude larger than the buoyancy flux estuarine MOC develops. A ratio ΔS∕ΔT > 1 indicates the (10−9 m2 s−3) in EXP1. Thus the Adriatic Sea results again “haline mode” of the marginal sea circulation with the shut to be an anti-estuarine basin characterized by a large wind down of deep convection and the possibility of an estuarine work energy source. MOC if the freshwater budget is negative. The collapse of deep convection is demonstrated to be possible also in the ΔS > 0.5 3.1.2 Theoretical marginal sea overturning estimates thermal mode case if ΔT . We obtain ΔS∕ΔT = 0.28 and 0.12 in EXP1 and EXP2 Spall (2011, 2012) analysed a marginal sea overturning respectively. Thus the Adriatic Sea is characterized by an circulation predicting that the freshwater input could stop anti-estuarine circulation with thermally driven deep water the anti-estuarine circulation of a theoretical basin. Rahm- formation processes despite a large runoff budget. Focus- storf (1995) speculates that increasing freshwater inflow in ing on the year 2002 which showed the largest river runoffs the Northern Atlantic may potentially reduce or even shut (Fig. 3), ΔS∕ΔT = 0.42 in EXP1 and 0.29 in EXP2, mean- down the overturning circulation. Cessi et al. (2014) show ing that the Adriatic deep water formation and the anti- that both buoyancy forcing and wind stress work are con- estuarine circulation characterize both experiments with nected to the strength of the circulation and thus, also to the and without river runoff but EXP1 is close to the collapse MOC developing between the marginal sea and the open of deep convection. ocean. Following Spall (2012) we assessed the non-dimensional thermal forcing parameter, µ/ε, and freshwater forc- 3.2 How is the intensity of Central Mediterranean ing parameter, γ/ε. The former describes the relative bal- MOC affected by runoff? ance between heat budget in the interior basin and the In order to better quantify the river influence on the CMOC, lateral eddy fluxes from the opening advecting warm water an inter-annual analysis of the meridional transport stream into the basin. The latter describes the relative balance function was carried out. between freshwater budget in the interior basin and the lat- The meridional transport stream-function, Ψ, is defined eral salt eddy fluxes at the opening again. In the Adriatic as (Pedlosky 1987): Sea lateral eddy fluxes are advecting heat and salt through x z Otranto thus opposing the net heat loss and water gain (see 1 Ψ(y, z) =− v(x, y, z)dxdz previous sections): thus it is important to evaluate these (2) nondimensional numbers to know how the lateral open x0 −H boundary forcings work against the surface buoyancy flux. Small values of these parameters with eventually negative with −H < z < 0 as the depth, x0 and x1 the more east- values of the freshwater parameter indicate that the lateral ern and more western sea points, v̄ is the time-averaged prevail over the surface cooling fliuxes in the basin interior meridional velocity. The velocity field is tangent to the iso- and may trigger the shutdown of deep convection and then pleths of Ψ, and positive Ψ values indicate anti-estuarine the reversal of the MOC. cells turning anti-cyclonically or clockwise, while nega- We computed the two parameters over the whole simu- tive values indicate estuarine cells turning cyclonically or lation period and discovered that the thermal parameter anticlockwise. is essentially the same in EXP1 and EXP2 experiments, Top panels of Fig. 8 show the transport stream-function 5 × 10−5 in EXP2 and 4.9 × 10−5 in EXP1, while the for EXP1 and EXP2, averaged over the whole simulation freshwater parameter is 7 × 10−4 in EXP2 and −2 × 10−2 period. A large anti-estuarine cell down to a 700–800 m in EXP1 (“Appendix 2” for details on the computation of depth is detected in both experiments in the Northern Ion- Spall’s coefficients). This means that the Adriatic Sea run- ian Sea and SAd sub-region, but with different intensi- off has the potential to shut down the deep convection and ties. Interestingly enough many estuarine cells exists in reduce the intensity of the antiestuarine CMOC. the domain: one at the surface in the NAd, one in the deep

1 3 River runoff influences on the Central Mediterranean overturning circulation

Fig. 8 Multiannual Meridional transport stream function for the Central Mediterranean Sea layers of the Southern Adriatic Pit around 41–42ºN, SAP, the same mechanism is present due to the complex topog- another in the Middle Adriatic Pit around 43ºN, MAP, and raphy and the slow abyssal cyclonic motion of deep water the last in the Northern Ionian abyss. The estuarine cells masses (Curchitser et al. 2001). attached to the seabed represent a more stagnant circula- Top panels of Fig. 8 demonstrate that with both a zero- tion in the Adriatic Pits and Ionian abyss. To note that the runoff and a realistic runoff case, the anti-estuarine char- estuarine cell which extends to the whole depth of the NAd acter of the CMOC prevails. Indeed, in EXP1, the second- is the consequence of a strong mixing of the entire water ary estuarine cells of the NAd, MAd, SAd sub-regions and column (i.e. about 35 m depth) due to heat losses and wind Northern Ionian basin are larger than in EXP2, however the stress. In the Ionian abyss the deep estuarine cell attached anti-estuarine MOC cell still dominates. to the seabed below the CMOC is linked to the SAd dense The estuarine component of the MOC may become waters overflowing the Otranto Strait and stratifying below more evident on a seasonal basis, particularly during sum- the CMOC cell (Curchitser et al. 2001; Bensi et al. 2013a). mer. Figure 9 focuses on summer 2002 and summer 2009 Deep estuarine cells exist below the wind driven anties- because these years had the largest spring river discharge tuarine MOC cells of the mid-to-deep water depths of (Fig. 3). In summer 2009 (Fig. 9, bottom panels), a well- the global ocean as demonstrated by Nikurashin and Val- defined surface estuarine cell characterizes the whole lis (2011) and De Lavergne et al. (2016). Recently Ferrari meridional extension of the basin with no differences et al. (2016) found that a deep overturning cell exists below between EXP1 and EXP2, which means that the wind the antiestuarine Atlantic MOC. This is due to the turbulent forcing becomes a dominant contribution to the estua- boundary layer very close to the sloping bottom of the deep rine secondary cells. During this season the wind work abyssal plains, generating diapycnal upwelling below the is maximum (Fig. 7, bottom panel) but the buoyancy flux Atlantic MOC. In the Mediterranean we can speculate that is still negative (Fig. 7, top panel) thus trying to force an

1 3 G. Verri et al.

Fig. 9 Summer 2002 (top panels) and Summer 2009 (bottom panels) Meridional Transport Stream Function for the Central Mediterranean Sea

anti-estuarine MOC in the Adriatic Sea. In summer 2002 Figure 10 focuses on year 2001 as we found that in 2001 (Fig. 9, top panels), the river influence is more significant the middle depth antiestuarine cell disappears with both due to the weaker wind forcing and the positive buoy- realistic (EXP1) and augmented runoff (EXP3 and EXP4). ancy forcing. In EXP1 the anti-estuarine MOC is weak This is due to the weak wind work (Fig. 7, bottom panel) and restricted to 200–400 m depths, while the secondary and the lowest heat budget (not shown). By comparing estuarine cells are stronger. However even in this case the Figs. 9 and 10 we then conclude that a large river runoff dominant overturning circulation is still anti-estuarine, in can produce a positive buoyancy flux without switching off agreement with the Spall’s (2012) freshwater and thermal the antiestuarine CMOC cell (as shown in Fig. 9 for year forcing parameters predictions. 2002) but a low heat flux and wind work with normal river Two additional experiments, EXP3 and EXP4, have runoff can in fact reverse it (as shown in Fig. 10 for year been performed by enhancing runoff of all rivers of 50 2001). and 100% (Table 1). The bottom panels of Fig. 8 show Overall by comparing the experiments with and without an abrupt weakening of the CMOC strength in EXP3 increased runoff, we demonstrate that rivers play a relevant and EXP4: the anti-estuarine cell does not disappear but role in the CMOC strength but they do not represent the reduces its intensity and extends only in the Northern Ion- dominant controlling mechanism. ian sub-basin up to 400 m depth, while the surface estua- Figure 11 provides the time series of the annual MTS rine cell is more pronounced and the deep estuarine cell of for all the experiments, by considering the averaged value the Ionian abyss enlarges over the SAd. Overall the multi- over 100–400 m depths in order to focus on the location of annual CMOC at middle depths remains an antiestuarine the anti-estuarine CMOC cell. These time series corrobo- cell in all the experiments. rate the previous result on river role in the CMOC strength.

1 3 River runoff influences on the Central Mediterranean overturning circulation

Fig. 10 Meridional transport stream function for the Central Mediterranean Sea over year 2001

Fig. 11 Time evolution of the Meridional transport stream function averaged over 100–400 m depths for all the experiments

As expected, EXP2 with zero-runoff shows the highest val- through the whole simulation period due to the dominant ues of the middle depth MTS and EXP4 the lowest ones. anti-estuarine CMOC cell, except for the years 2001 (along The EXP4 retains positive values of the middle depth MTS with EXP1 and EXP3, as shown in Fig. 10) and 2003.

1 3 G. Verri et al.

As far as we know an analysis of the MOC for the East- DW formation in winter 2000–2001 and the lowering in ern Mediterranean (EMOC) has been performed only by 2007 looking at the NAd sub-region, the absence of dense Pisacane et al. 2006 and Amitai et al. (2016). Pisacane water over 2000–2003 in the MAd and till 2005 in the SAd (Pisacane et al. 2006) EMOC cell resembles our CMOC which is known to be linked to reduced MLIW inflow and cell intensity and extension with a well defined anties- high temperature; the high dense water formation in 2006. tuarine character but this cell is computed over a different There is a perfect agreement with DW volumes found by model domain and the interannual average is shown only Gunduz et al. (2013) and the SAd DW formation fits the for the summer time. Amitai et al. (2016) evaluated the observations by Cardin et al. (2011). Our results also con- MOC in the Adriatic basin: their maximum value of 0.5 Sv firm the exceptional dense water formation in winter 2012 is comparable with our results, which show a maximum as consequence of an extreme wintertime cooling event, value of about 0.3 Sv at 300 m depth close to the Otranto which produces DW volumes capable to saturate the vol- Strait (Fig. 8, left panel). However Amitai et al. (2016) dif- ume of NAd, i.e. 2357 km3, and MAd, i.e. 4039 km3, sub- fers because the antiestuarine cell is restricted only to the regions (Mihanovic ́ et al. 2013; Janekovic ́ et al. 2014). On southern part of the SAd Pit while an additional estuarine the other hand our estimated SAd DW volumes are pretty cell occupies its northern part. higher with respect to Oddo and Guarnieri (2011). In our full dynamics experiment, EXP1, we found the transport 3.3 How do rivers influence the formation of dense of NAd dense water toward the middle Adriatic had a peak water in the Adriatic Sea? value of 1.6 Sv in 2012 while 0.6 Sv is found by Janek- ovic ́ et al. (2014), actually defined over a smaller region The aim here is to establish how rivers impact the dense thus cannot be directly compared. Our SAd DW formation water formation processes in the Southern Adriatic Sea and annual rate reaches the peak of 1.52 Sv in 2006, higher thus impact the CMOC. The conditioning factors of open- then 0.64 Sv estimate by Oddo and Guarnieri (2011) but sea convection and dense water formation in the South- our estimate is consistent with the known potential peaks ern Adriatic are known to be the permanent cyclonic gyre of SAd DW rates, following Mantziafou and Lascaratos sustained by wind stress curl, the large buoyancy losses, (2008) and Vilibic and Supic (2005). mainly known to be due to large heat losses, and the inflow We found EXP2, without river forcing, shows 20–30% of LIW through the Otranto Strait. Moreover the SAd deep larger SAd dense water volumes than EXP1. Previous stud- layers are partially filled by the NAd dense water which ies suggest that rivers affect SAd dense water volumes partially the MAd Pit and partially flows southward over because they reduce the lateral advection of NAd dense the Italian western shelf and reaches the Bari Canyon, usu- waters which are known to flow along the western shelf ally 2 months after its generation. Here the boundary cur- and slide down into the Southern Adriatic Pit near the Bari rent of NAd dense water mixes turbulently with adjacent canyon (Vilibić and Orlić 2001–2002; Mantziafou and Las- warmer and less saline waters and sinks to the Southern caratos 2004, 2008; Wang et al. 2007). However, we found Adriatic Pit (Vilibić and Orlić 2001; Bensi et al. 2013b). this mechanism is not sufficient to account for EXP1 and Figure 12 shows the dense water volume formed in the EXP2 SAd dense water volume difference because the lat- NAd, MAd and SAd sub-regions computed as the water ter is larger than the sum of NAd and MAd dense water volume with larger potential density anomaly than the volume differences between EXP1 and EXP2. This means threshold value, 29.2 kg m−3. that rivers affect SAd dense water volumes not only by In the NAd, where the major discharge is concentrated, reducing the lateral advection of NAd dense water but they buoyancy gains due to river runoff oppose the precondition- also oppose the local preconditioning factors of the dense ing factors of DW formation i.e. the winter surface cool- water formation in the SAd, i.e. the “open sea convection” ing, the outbreaks of cold and dry wind like Bora and the mechanism, as they locally change the vertical stratification autumn/winter NAd cyclonic gyre (Fig. 12, top panel). Riv- characteristics. ers may affect DW volume also in the MAd as they reduce To explain the river impact on the open sea convection, the DW advected from the NAd (Fig. 12, middle panel). Fig. 13 shows seasonal θ–S diagrams in three Adriatic sub- Looking at the full dynamics experiment (EXP1) we region zonal sections for winter 2009. The main difference find that the mean DWF rate of 0.3 Sv is consistent with between EXP1 and EXP2 θ–S diagrams is the degree of the literature estimates (Artegiani et al. 1989, 1997a, b; stratification of the water column: river runoff determines Lascaratos 1993; Cushman-Roisin et al. 2002; Curchitser a well stratified water column in all the three subregions et al. 2001; Manca et al. 2002; Mantziafou and Lascaratos and different water types with characteristic θ and S values 2004, 2008; Pinardi et al. 2015). The interannual forma- can be distinguished. On the other hand the no-river case, tion rates match the recent estimates by Oddo and Guarni- EXP2, shows an almost constant salinity value in the three eri (2011) and by Gunduz et al. (2013): the collapse of zonal sections.

1 3 River runoff influences on the Central Mediterranean overturning circulation

Fig. 12 Daily averaged time series of water volumes with 29.2 −3 𝜎𝜃 > kg m in the NAd (top panel), MAd (middle panel) and SAd (bottom panel) sub-regions. The blue line stands for EXP1, the red line for EXP2

Thus we conclude that river runoff primarily influences 3.4 How do rivers influence the Otranto water the dense water formation processes by changing the SAd exchange and the volume of Adriatic dense water vertical stratification characteristics. However these stratifi- that spreads into the Northern Ionian Sea? cation changes can determine changes in the CMOC characteristics only if the overall buoyancy forcing is positive, Figure 14 shows the Otranto inflow-outflow regime for as shown for the 2002 case. In the 2009 case, wind stress EXP1 and EXP2. EXP2 shows an “horizontally detached” work is large and buoyancy forcing still negative, thus bal- exchange flow with inflowing water on the eastern side and ancing the large river runoff stratification effects and pro- outflowing water on the western one. EXP1 shows a weaker ducing no relevant changes in the CMOC. exchange pattern with lower meridional velocity through

1 3 G. Verri et al.

16 26.7 all depths and strong exchange flux narrowed to the upper levels. 28.8 27.9 We know that the Modified Levantine Intermediate 29 14 27 Water, MLIW, arriving from the Rhodes area of formation 28.2 occasionally enters the Adriatic Sea mainly in summer and 27.3 27.6 autumn between 150 and 400 m (Malanotte-Rizzoli et al. C) °

( 12 1997; Robinson et al. 2001; Manca et al. 2002; Mantziafou

θ and Lascaratos 2004, 2008). Moreover it is well known from the literature that the SAd dense water overflows the 29.7 10 3 Otranto Strait mainly along its western part (Curchitser

28.5 et al. 2001). 29.4 Our modeling findings prove for the first time that rivers 8 reduce both the MLIW inflow on the Eastern shelf and the 36 37 38 39 S (psu) SAd DW outflow on the western side. The MLIW inflow is not systematic but depends on the

16 26.7 27.9 cyclonic/anticyclonic character of the Ionian intermediate water circulation: our experiment covers most of the 28.2 27 cyclonic decadal phase of the Ionian near-surface circula- 14 tion, i.e. 1997–2007 (Gacic ́ et al. 2010; Bensi et al. 2013a, 28.5 b), which favors the entrance of MLIW into the Adriatic 27.3 basin. 29.1 ° C)

( 12 Timeseries of the meridional heat and volume trans-

θ ports at the Otranto Strait are provided in Fig. 15. Rivers 29.4 are shown to reduce the net ingoing heat transport (Fig. 15, 27.6 28.8 10 29.7 upper panel) and to increase the outgoing volume trans- 30 port (Fig. 15, bottom panel). The annual values of heat and volume transports we found are fully consistent with the 8 annual averaged heat gain, i.e. 2.92 TW, and volume out- 36 37 38 39 flow, i.e. −0.003 Sv, computed by Mantziafou and Lascara- S (psu) tos (2004). 16 Bottom panel of Fig. 15 confirms river role on the vol- 26.7 MLIW 28 ume exchange at the Otranto Strait as already stressed in 28.5 Fig. 14. Rivers increase the freshwater outflow at the sur- WACC 14 27 29 face, they reduce the inflow of salt and warm water on the 27.3 27.9 eastern flank at the surface, i.e. the Ionian Surface Water (ISW), and middle depths, i.e. the MLIW; finally rivers ESAC C)

° 12 reduce the outflow of Adriatic DW through the bottom

( Western

θ 27.6 28.2 Bottom (Cushman-Roisin et al. 2002), which represents only 30% current of the total outflow. Thus the overall effect is an increase

10 29.7 of the total outflow at the Otranto Strait. Years 2005 and 29.4 30 2006 are the only ones with no differences in terms of volume transport between the two EXPs, because other forc- 8 ings (e.g. surface heat losses, ISW and MLIW inflow) are 36 37 38 39 prevailing on river freshwater forcing. S (psu) The Adriatic dense waters outflow the Otranto Strait and spread into the abyssal Ionian Sea, generally occupying the Fig. 13 θ-S diagram in winter 2009 for EXP1 (blue dots) and layer below the salt waters coming from the Cretan Sea red dots EXP2 ( ) in three zonal sections related to NAd sub-region and the Levantine basin (Roussenov et al. 2001; Curchitser at 44.38°N (top panel), MAd at 42.6°N (middle panel) and SAd at 41.61ºN (bottom panel), respectively. The Western Adriatic Coastal et al. 2001; Bensi et al. 2013a). Current, WACC, the Eastern Southern Adriatic Current, ESAC, the Figure 16 shows the seasonal potential density anomaly Western bottom current and the Modified Levantine Intermedi- of the 200 m layer above the seabed in both EXP1 and bottom panel ate Water, MLIW, are marked in the according to the EXP2 and their differences with a zoom on the Ionian Sea. known range of temperatures and salinities The maps show spring 2012 because this was the year with

1 3 River runoff influences on the Central Mediterranean overturning circulation

Fig. 14 Transect of the meridional velocity at the Otranto channel (i.e. 40°N). Firm lines mean inflow,dashed lines mean outflow, the whileline stands for zero meridional velocity

1 3 G. Verri et al.

Fig. 15 Upper panel Annual time series of heat transport (units are TW) at the Otranto Strait. Bottom panel Annual timeseries of meridional volume transport (units are Sv) at the Otranto Strait. Positive values mean northward or ingoing transport, negative values southward or outgoing transport

one of the largest dense water volume formation (as shown toward the Ionian Seaia. This suggests that river runoff has in Fig. 12). consequences on the Northern Ionian Sea water mass struc- SAd dense water initially spreads into the Northern Ion- ture and abyssal mixing processes. It is interesting to con- ian Sea following the topography. The Adriatic outflowing nect such outflow differences with the abyssal estuarine cell dense water is known to move mainly along the isobaths on of Fig. 8: EXP1 shows a stronger deep estuarine cell than the Italian shelf at about 600–1000 m in almost geostrophic EXP2, expecially on seasonal basis. We speculate the slow balance (Budillon et al. 2010), a secondary branch follows a abyssal circulation of deep water masses in EXP1, as rivers meridional path driven by the local bathymetry (Hainbucher riduce the SAd DW outflow to narrower bands, as well as et al. 2006); both branches slowed down and sink due to fric- the complex topography promote the diapycnal upwelling tion and turbulent mixing with the ambient waters, thereby and result in a stronger abyssal estuarine cell. creating a nearly homogeneous layer below 1200 m (Curchit- ser et al. 2001). Wu and Haines (1996) found newly formed Adriatic dense waters fill the Ionian abyss in 2–3 years. 4 Conclusions and future developments Figure 16 shows that the bottom boundary current forced by dense water outflow from the Otranto Strait is charac- In this paper we investigated the influence of river terized in EXP1 by less dense waters that intrude offshore freshwater inflow on the vertical overturning cell of the with respect to EXP2 as expected. Furthermore the Adri- Central Mediterranean Sea (CMOC). A twin experi- atic dense water outflow in EXP2 is confined to a narrower ment, with and without runoff, was carried out from the band against the Italian shelf with less lateral spreading beginning of 1999 to the end of 2012. For EXP1 with

1 3 River runoff influences on the Central Mediterranean overturning circulation

estimate of river runoff and then we subtract the runoff to understand its impacts. The anti-estuarine CMOC has its downwelling branch in the SAd deep water formation areas and thus we analyze first the marginal sea overturning circulation characteristics to understand differences due to runoff. We first compute the surface forcing budget and the theoretical marginal sea estuarine/antiestuarine parameters devised by Spall (2012). The long term and surface average of the buoyancy forcing over the Adriatic Sea is negative but during years of large runoff it changes sign, thus supporting the conjecture that the Adriatic Sea could change from anti-estuarine to estuarine vertical circulation. However the computation of Spall (2012) thermal and freshwater nondimensional parameters demonstrates that river runoff cannot reverse the dominant anti-estuarine character of Adriatic circulation or shut down the deep convection in the basin interior because the system is mainly in the thermal mode and wind work is always positive. Our results for the realistic runoff case show that the multiannual CMOC for the decade 2000–2012 is a permanent anti-estuarine meridional overturning cell, occupying the Southern Adriatic and the Ionian Sea, plus secondary estuarine cells in the NAd and in the MAd, SAd deep layers as well as in the Northern Ionian abyss. The deep estuarine cell in the Ionian abyss is pointed out for the first time in this study and we speculate this is the result of diapycnal upwelling in the turbulent boundary layer close to the sloping bottom of the Ionian abyss. A key result is that the CMOC is largely driven by wind work and heat fluxes but large and anomalous river runoff can affect its strength, enhancing the amplitude of the secondary estuarine cells and reducing the intensity of the dominant anti-estuarine cell. Two additional experiments have been performed by enhancing the realistic runoff estimate of all rivers of 50% and 100%. Both experiments show an abrupt reduction of the antiestuarine CMOC cell on multiannual basis: the antiestuarine cell does not disappear but strongly reduces its intensity and extends only in the Northern Ionian sub- basin, while the deep estuarine cell of the Ionian abyss Fig. 16 Seasonal potential density anomaly on a 200 m layer above enlarges over the SAd. seabed and difference between EXP1 and EXP2 with zoom on the We found that on 2001 the middle depth antiestuarine Ionian Sea cell disappears with both realistic and augmented runoff, due to the weak wind work and the lowest heat budget. This realistic runoff estimate, the comparison with satellite means that rivers play a relevant role in the CMOC strength SST and ARGO profiles shows that the model is capa- but they do not represent its dominant forcing mechanism ble to reproduce the major characteristics of the observed and the potential role of river runoff on the intensity and water masses. We then take this experiment to be the best direction of the CMOC has to be considered jointly with estimate of reality and we subtract the river runoff from wind work and heat flux, as they largely contribute to the the simulation and we study the differences between the energy budget of the basin. two experiments. Our study is “mechanistic”, i.e. we use We focused on the downwelling branch of the Central a full forcing and dynamics simulation with a realistic Mediterranean MOC, which develops in the Adriatic basin

1 3 G. Verri et al.

v v v v 1 p 2v 2v 2v due to dense water formation processes. Rivers are demon- + u + v + w =− + A + + K − fu t x y z y m x2 y2 m z2 strated to affect the Adriatic dense water volumes. Previous 0 studies showed that rivers reduce the dense water formation (4) in the Northern sub-region where most discharge is located. p =−g Here we show that rivers also directly affect the vertical z (5) mixing processes in the Southern Adriatic sub-region by changing the water column stratification in the SAd and ∇ ⋅ u⃗ = 0 (6) thus decreasing the dense water volumes. Finally we showed that the Adriatic dense waters over- 𝜕𝜂 +∇ ⋅ u��⃗ (H + 𝜂) = P + R − E flowing the Otranto Strait are less dense in a realistic runoff 𝜕t H Hbaro (7) regime and confined to a narrower band against the Ital- ian shelf with less lateral spreading toward the Ionian Sea 𝜕S 𝜕2S +∇⋅ uS⃗ = A ∇2 S + K center. 𝜕t t H t 𝜕z2 (8) Future investigations will point to an advanced implementation of river discharge in the model and the extension 𝜕𝜃 𝜕2𝜃 +∇⋅ u⃗𝜃 = A ∇2 𝜃 + K to the whole Mediterranean Sea. The next questions could 𝜕t t H t 𝜕z2 (9) consider the connection between the CMOC and the zonal overtuning cell of the Mediterranean Sea that supports the = (, S, p) (10) flow of LIW in the different deep water formation areas of The first two Eqs. (3) and (4) are the Navier–Stokes the eastern and western Mediterranean Sea. equations with the Boussinesq approximation for the Acknowledgements We thank to the anonymous reviewers for their three dimensional horizontal velocity vector field very useful comments and discussions. u��⃗H = (u, v) while Eq. (5) is the hydrostatic approximation of the vertical velocity component. The coefficients Am Funding This work was funded by the Italian Project TESSA and Km are the momentum eddy coefficients for horizon- through Centro Euro-Mediterraneo sui Cambiamenti Climatici, tal and vertical mixing respectively, fv and −fu are the Lecce, Italy. Nadia Pinardi was funded by the GEMINA Project at CMCC and Paolo Oddo by the RITMARE Project at INGV. horizontal components of Coriolis term. The Eq. (6) is the continuity equation with the incompressible approxi- Open Access This article is distributed under the terms of the mation u⃗ = (u, v, w) which allows to compute the vertical Creative Commons Attribution 4.0 International License (http:// velocity, w, as diagnostic variable. The Eq. (7) is the creativecommons.org/licenses/by/4.0/), which permits unrestricted vertically integrated continuity equation, written as a use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a prognostic equation for the free surface starting from the link to the Creative Commons license, and indicate if changes were vertical integration of Eq. (6) and replacing the defini- 1 𝜂 made. u��⃗ = u��⃗dz tion of barotropic velocity that is Hbaro 𝜂+H ∫−H H . The Eqs. (8) and (9) are the advection/diffusion equa- A K Appendix 1: The numerical model configuration tions for tracers with t and t the horizontal and vertical mixing coefficients of tracers. The numerical simulations were carried out using three- Finally the sea state Eq. (10) prescribes the ocean dimensional, finite difference primitive equations Nucleus water density as a non linear empirical function of poten- for European Modelling of the Ocean code, NEMO v 3.4 tial temperature, salinity and pressure (following Jackett (Madec 2008). and McDougall 1995). The model solves prognostic equations for potential tem- The sea surface height Eq. (7) and the associated baro- perature, practical salinity, horizontal velocity components tropic velocity equations are solved by the time-splitting in the meridional and zonal directions, sea surface height formulation, thus using a smaller time step than for three- and diagnostic equations for vertical velocity, hydrostatic dimensional prognostic variables. pressure and potential density. In order to solve the mesoscale variability of the Adri- Boussinesq and hydrostatic hypotheses are assumed. atic Sea, at least in the Southern sub-basin, a horizontal grid resolution equal to 1/45° was chosen, corresponding 2 2 u u u u 1 p u u to 2.47 km in the meridional direction and 1.72–2.13 km + u + v + w =− + Am 2 + 2 t x y z 0 x x y in the zonal direction. The literature shows the first baro- 2u clinic Rossby radius of deformation in the Mediterra- + K + fv m z2 nean Sea is around 10–12 km (Grilli and Pinardi 1998; (3) Pinardi and Masetti 2000) if we take the open flow scale

1 3 River runoff influences on the Central Mediterranean overturning circulation

C < 0 variables, but the local values may significantly reduce For inward propagation, 𝜙x , the algorithm pre- C = C = 0 depending on season and latitude and moving towards the scribes x y thus reduced to a relaxation shelf areas. In the Northern Adriatic Sea it reduces up to condition. about 3–5 km in summer and 1 km in winter (Paschini For the horizontal velocity components, u3d and v3d, et al. 1993; Masina and Pinardi 1994; Bergamasco et al. the imposition scheme is used and thus, the incoming and 1996). This means our model can explicitly resolve the outgoing information is totally determined by the coarser mesoscale activities in the Adriatic Sea, at least in the model data, irrespective of the inner solution. Southern sub-basin, on seasonal as well as on interan- In addition, the horizontal velocity component normal nual basis with the only exception of the Northern Adri- to each boundary is uniformly adjusted according to the atic where the model may result eddy-permitting but not “interpolation constraint” procedure (Pinardi et al. 2003) in eddy-resolving. order to preserve the total volume transport after data inter- The model bathymetry, covering both the Adriatic and polation from the coarse to the fine resolution grid. Ionia Sea, is taken from the US Navy 1/60° bathymetric For the barotropic velocities, uBT and vBT, Flather’s database DBDB1 using bilinear interpolation. scheme (1976) was adopted. The barotropic velocity com- A total of 121 unevenly spaced z-levels with par- ponent normal to the eastern and western boundaries is tial steps were adopted in the vertical direction. Partial given by Flather’s equation: steps allow a better representation of the bathymetry. The higher resolution in the top layers (23 levels in the gH u = u − − BT BTnested nested (12) top 35 m which is the mean depth of the NAd subregion) √H leads to an improved simulation of the bottom flow in the � � where nested is the coarser model sea surface height at the NAd and vertical mixing during higher stratification in the summer. boundary interpolated over the finer model grid, is the finer model sea surface height at the boundary andu BT is There are two open boundaries on the eastern and west- nested ern sides of the model domain. Open boundaries data are the coarser model normal barotropic velocity over the finer provided as monthly means and involve the following model grid computed as prognostic variables interpolated on the model grid: zonal u v 1 velocity ( 3d), meridional velocity ( 3d), potential tempera- uBT = u3d dz nested H + nested ture ( ), salinity (S), and the sea surface height ( ). −H For both the lateral open boundary conditions, LOBCs, and the initial conditions, ICs, data are taken The tangential barotropic velocity is set equal to zero: from daily analysis of the operational Mediterranean vBT = 0. forecasting System, MFS (Tonani et al. 2008; Oddo et al. In Flather’s formula, values at the boundary follow a 2009; Pinardi and Coppini 2010) based on the same code, “zero gradient boundary condition” which means B = B−1 NEMO, and covering the whole Mediterranean basin (subscript B stands for boundary line values). This avoids with 1/16 horizontal resolution. numerical instabilities. The numerical schemes adopted for the LOBCs are The bottom boundary condition is applied only on described below. momentum and consists of a quadratic friction. Marchesiello’s algorithm (Marchesiello et al. 2001) No slip boundary conditions are adopted along the was used for active tracers. It consists of the 2D radiation coastline for tangential velocity. condition plus a relaxation/nudging term as follows: In order to define the air-sea interaction, the vertical fluxes 1 of momentum, heat and salt and the vertical velocity were + C + C =− − t x x y y nested (11) parameterized at the sea surface. These parameterizations are the surface boundary conditions (SBCs) of the model. Wind where φ is the tracer ( or S), nested is the coarser model stress and heat flux components are computed by means of (MFS) solution for the tracer interpolated on our model “bulk formulae” (Castellari et al. 1998; Maggiore et al. 1998; grid and provided monthly. The time scale for the nudging Oddo and Guarnieri 2011; Madec 2008) using atmospheric term, τ, is constant and equal to 1 day for inward propaga- data provided by the European Centre for Medium Weather tion and 15 days for outward propagation. Forecasts (ECMWF). These atmospheric data (2 m air tem- C > 0 C For outward propagation, i.e. 𝜙x where x is the perature, 2 m dew point temperature, total cloud cover, mean component of the phase velocity normal to the boundary, sea level atmospheric pressure, meridional and zonal 10 m C = 0. the tangential component is set equal to zero, y wind components) are operational analyses with a 6 h frequency and with 0.5° horizontal resolution up to 2008 and 0.25° thereafter. It should be noted that the increase in spatial

1 3 G. Verri et al. resolution of the ECMWF fields could lead to a spurious The wind stress involved in the surface boundary con- increase of the wind stress magnitude. dition for momentum is calculated from the relative winds Only the precipitation rate (P) data are extracted from the with the formula: CMAP (CPC, Climate Prediction Center, Merged Analysis of Tw = 0aCD Urel Urel Precipitation) monthly data set with a horizontal resolution (18) U u u u v of 2.5° × 2.5°. This coarse precipitation data set is a weak- where rel = w − s =( rel, rel) is the relative wind field u ness of our model formulation but precipitation data set from that is the 10 m wind horizontal velocity w subtracted from u ECMWF was still too poor for the first decade of the XXI the sea surface horizontal velocity S, 0a is the density C T T u century (Haiden et al. 2015) and this would have offset the of the moist air and D a, s, w is the drag coefficient buoyancy budget in an unphysical manner. which depends on air temperature, sea surface temperature The surface boundary condition for temperature involves and wind amplitude according to Hellerman and Rosen- a balance between solar short-wave radiation Qs(computed stein (1983). using Reed’s formula, 1977), long-wave radiation Ql (com- The momentum boundary condition at the surface is: puted using Bignami et al. 1995), latent Qe and sensible Qh (u, v) 0Km = wx, wy heat fluxes [by means of bulk formulae proposed by Kondo z z= (19) (1975)]. Reed’s formula is: where wx = 0aCD Urel urel and wy = 0aCD Urel vrel are the wind stress components and Km is the vertical mixing Qs = Qtot(1 − 0.62C + 0.0019 )(1 − ) (13) coefficient for momentum. C where Qtot is the clear-sky radiation, is the fractional The freshwater balance defined as evaporation minus cloud cover, is the noon sun altitude in degrees, and is precipitation and runoff (with the latter divided by the cell the sea surface albedo. The albedo is computed as a func- area of the river mouth), E–P–R/A, is directly involved in tion of the sun zenith angle for each grid point from Payne the surface boundary conditions for salinity and for vertical (1972). velocity. The evaporation rate (E) is calculated by the latent Q The Bignami formula is: heat flux according to E = e. Le Q = T4 − T4 0.653 + 0.00535e 1 + 0.1762C2 l S A A The salinity boundary condition at the surface reads: (14) S K = S (E − P − R∕A) where is the ocean emissivity, is the Stefan Boltzmann t z z= (20) e T z= constant, A is the atmospheric vapor pressure, S is the sea T surface temperature predicted by model, A is the 2 m-air where is the sea surface elevation and Sz= is the ocean temperature. model surface salinity except prescribed ad-hoc salt values The sensible Qh and latent Qe heat fluxes are parameter- at river mouths. ized through the Kondo bulk formula: The surface boundary condition for the vertical velocity is as follows: Q = C C v T − T h A p H S A (15) w − − u ⋅ ∇ = (E − P − R∕A) z= t z= H (21) Qe = ALeCE v esat TS − resatTA 0.622∕pA (16) where w is the vertical velocity. where A = A p, TA, r is moist air density, r is the rela- With regard to the dynamics, the following choices were tive umidity, C p is the specific heat capacity at constant selected: vector invariant form for momentum advection, pressure, CH and CE are the turbulent exchange coefficients bi-Laplacian operator for lateral diffusion and horizontal computed according to Kondo (1975), Le is the latent heat v eddy viscosity coefficient equal −5 × 107 m4 s−1 according of vaporization, esat is the vapor pressure, is the wind to a tuning procedure starting with MFS values, implicit speed modulus, and pA is the atmospheric pressure. vertical diffusion and TKE turbulence closure scheme For the heat flux boundary condition at the surface, we (Mellor and Blumberg 2004) to provide the vertical eddy assume: coefficients. 1 K = 1 − T Q − Q − Q − Q With regard to the active tracers: MUSCL advection 0 t r s l h e (17) z z= Cp scheme, bi-Laplacian operator for lateral diffusion and hor- 7 4 −1 izontal eddy diffusivity coefficient equal to −3 × 10 m s where T r is the Jerlov (1976) transmission coefficient for a K according to a tuning procedure, implicit and TKE depend- “clear” water type and t is the vertical mixing coefficient ent vertical diffusion. for traces.

1 3 River runoff influences on the Central Mediterranean overturning circulation

Appendix 2: The computation of Spall parameters The non-dimensional parameter = cP∕L is the ratio of for semi‑enclosed seas the heat flux toward the basin interior due to lateral eddies compared to that advected into the Adriatic Sea through the Following Spall’s studies (2004, 2010, 2011, 2012) on inflowing boundary current along the Southern Adriatic overturning circulation in marginal seas, we computed two eastern shelf. The inflowing boundary current is assumed non-dimensional coefficients which represent the relative to be a geostrophic current in thermal wind balance. The balance between heat and freshwater budget in the interior value of is very small for stable boundary currents and of the Adriatic Sea due to air–sea interaction and the lateral increases for boundary currents that are sufficiently unsta- eddy fluxes that advect warm and salty water in the basin ble that they lose all their heat to the interior of the basin interior and are detach from the cyclonic boundary current before it is carried all the way around the marginal sea. which inflows along the eastern side from the open ocean Moreover the thermal and freshwater forcing parameters ∗ and encircles the marginal sea. required to compute T , that is the difference between the The combinations ∕ and ∕ are called respectively inflowing temperature and the temperature of the atmos- thermal and freshwater forcing parameter and are described phere over the interior of the marginal sea, as follows : below: 2.56 (◦C) in EXP1 T∗ = T − T = 1 A 2.84 (◦C) in EXP2 AΓf0 cP = 2 ∗ L (22) T GCpH T where T1 is the mean temperature of the inflowing current along the eastern boundary derived from the EXPs, TA is the mean 2 m temperature over the Adriatic basin extracted 8A f S (E − P − R) cP = 0 0 0 S from ECMWF 25 km dataset. 2 2 ∗2 L (23) gH T T Finally the surface freshwater flux is defined as follows: E − P = 0.66 × 10−7 (ms−1) in EXP2 where A is the area of the Adriatic sea surface (from model domain), is the relaxation constant for the basin sea sur- E − P − R =−2.18 × 10−7 ms−1 in EXP1 face temperature toward the atmospheric temperature (from All the quantities described above enable to compute the Spall 2011), f0 is the Coriolis parameter, T is the thermal thermal and freshwater forcing parameters in Eqs. (22) and expansion coefficient (from Cessi et al.2014 ), S is the (23), giving: haline expansion coefficient (from Cessi et al.2014 ), H is P the depth of the sill (from model domain), is the perim- 5.0 × 10−5 in EXP1 c ∕ = eter of the basin interior (from model domain), P is the 4.9 × 10−5 in EXP2 thermal capacity (from Cessi et al. 2014), L is the L is the width of the sloping topography over which the inflowing −2.0 × 10−2 in EXP1 boundary current lies (thus computed from model results ∕ = 7 × 10−4 in EXP2 as the cross-shore width of the inflowing boundary current along the eastern shelf of the Adriatic basin). hx As discussed by Spall (2011), ∕ is a measure of the The variable 𝛿 = =−0.33 represents the topog- − ̄𝜌 x∕ ̄𝜌 z relative influence of lateral eddy heat fluxes from the raphy slope over the mean isopycnal slope in the bound- boundary current into the basin interior compared to heat ary current, thus both computed along the Southern Adri- loss to the atmosphere. For 𝜇∕𝜀≪1, lateral eddy heat flux atic eastern shelf (x-direction stands for zonal direction from the boundary is very strong and leads to a relatively and z means depth). To note that δ has been computed by warm basin interior so that T ≈ T1, if 𝜇∕𝜀>1 the bound- considering a zonal transect of potential density anomaly ary current is relatively stable and the atmosphere is able at 40.8ºN (so just north of the Otranto Strait) on annual to strongly cool the basin interior so that T ≈ TA. Similarly basis and focusing on the eastern side of the basin. For ∕ describes the relative role of surface forcing and lat- cyclonic boundary current, δ < 0 and the topography acts eral eddy fluxes in the salinity balance. Large values of ∕ to stabilize the boundary current and reduce the amount indicate dominance of atmospheric forcing implying fresh- of lateral eddy flux into the interior. The quantity water gains in the basin interior that are not balanced by c 2 = 0.025e = 0.05 is an efficiency coefficient that lateral eddy fluxes of salt from the boudary current, and depends on the bottom slope and regulates the eddy heat small values indicate strong lateral eddy fluxes. flux from the boundary current into the interior (Spall In order to evaluate the shutdown of deep convection 2004). and the reversal of the overturning circulation, the temperature and salinity anomalies of the basin convective

1 3 G. Verri et al.

Table 3 Summary of the key parameters of the Twin experiment The theoretical limit for shutdown of deep convection is ΔS∕ΔT ⩾ 0.5 Parameter EXP1 EXP2 , thus possible also in the thermal mode. According to our findings, both EXPs are in the thermal −5 −5 ∕ 5.0 × 10 4.9 × 10 mode and EXP1 is closer to the threshold limit for shut- −2 −4 ∕ −2 × 10 +7 × 10 down of deep convection than EXP2. T Δ 0.35 0.28 Results collected for EXP1 and EXP2 are summarizes S Δ 0.10 0.03 in Table 3 and show that deep convection in the Southern S T Δ ∕Δ 0.28 0.12 Adriatic, surface cyclonic boundary current and anti-estua- The model-diagnosed quantities are the thermal forcing parameter rine overturning circulation of the Adriatic basin character- ∕, the freshwater forcing parameter ∕, the temperature anomaly ize both experiments but in EXP1, with a realistic param- of the convective water mass ΔT and the salinity anomaly of the con- eterization of river runoff, the freshwater forcing coefficient ΔS vective water mass is negative and the ratio ΔS∕ΔT close to 0.5. This cor- roborates strong river discharge in the Adriatic Sea has the water mass have been computed as normalized differ- potential to trigger the shutdown of deep convection and ences of T (i.e. ΔT) and S (i.e. ΔS) between basin interior the weakening of the anti-estuarine overturning circulation. and boundary currents (Spall 2012). A set of two non-dimensional equations has been derived to compute ΔT and ΔS (Spall 2012), these equa- References tions include the non-dimensional parameters ∕ and ∕ and describe ΔTand ΔS as function of basin geom- Amitai Y, Ashkenazy Y, Gildor H (2016) Multiple equilibria and etry, atmospheric forcing and lateral eddy fluxes. overturning variability of the Aegean-Adriatic Seas. Glob Planet Change The simplified formula suggested in Spall (2012) are: Artegiani A, Azzolini R, Salusti E (1989) On the dense water in the ΔT = T − T∕T∗ Adriatic Sea. Oceanol Acta 12:151–160 1 (24) Artegiani A, Paschini E, Russo A, Bregant D, Raicich F, Pinardi ∗ N (1997a) The Adriatic Sea general circulation. Part I: air– ΔS = S1 − S S∕ T T (25) sea interactions and water mass structure. J Phys Ocean 27:1492–1514 where T and S for the basin interior and T1 andS1 for the Artegiani A, Paschini E, Russo A, Bregant D, Raicich F, Pinardi N inflowing current have been computed over 50–200 m (1997b) The Adriatic Sea general circulation. Part II: baroclinic depth. circulation structure. J Phys Ocean 27:1515–1532 The ratio ΔS∕ΔT < 1 means the stable circulation Bensi M, Rubino A, Cardin V, Hainbucher D, Mancero-Mosquera I state is in the “Thermal mode” with surface heat losses (2013a) Structure and variability of the abyssal water masses in the Ionian Sea in the period 2003–2010. J Geophys Res Oceans and freshwater gains prevailing the lateral eddy advection 118(2):931–943 of warm and salt water in the basin interior (thermal and Bensi M, Cardin V, Rubino A, Notarstefano G, Poulain PM freshwater forcing coefficients are significantly high). In (2013b) Effects of winter convection on the deep layer of this case the density contrast is dominated by the temper- the Southern Adriatic Sea in 2012. J Geophys Res Oceans 118(11):6064–6075 ature difference and the water in the interior of the mar- Bergamasco A, Gacic M, Boscolo R, Umgiesser G (1996) Winter ginal sea is more dense than that in the boundary current. oceanographic conditions and water mass balance in the North- If ΔS∕ΔT > 1, the stable circulation state is in the ern Adriatic (February 1993). J Mar Sys 7:67–94 “Haline mode” with surface heat and freshwater budget Bethoux JP (1979) Budgets of the Mediterranean Sea: their depend- ence on the local climate and on the characteristics of the Atlan- of the interios basin favoring the lateral eddy advection tic waters. Oceanol Acta 2:157–163 of warm and salt water (thermal and freshwater forcing Beuvier J, Sevault F, Herrmann M, Kontoyiannis H, Ludwig W, coefficients are low enough and the latter is eventually Rixen M, Stanev E, Béranger K, Somot S (2010) Modelling the negative). In this case the density contrast is dominated Mediterranean Sea interannual variability over the last 40 years: focus on the EMT. JGR-Ocean 115(C8):1978–2012 by the salinity difference and the water in the interior Bignami F, Marullo S, Santoleri R, Schiano ME (1995) Longwave of the marginal sea is less dense than that in the bound- radiation budget in the Mediterranean Sea. J Geophys Res ary current. Thus the boundary corrent detaches from 100(C2):2501–2514 the eastern shelf and spreads in the interior basin. This Budillon G, Bue NL, Siena G, Spezie G (2010) Hydrographic characteristics of water masses and circulation in the Northern Ionian means in haline mode the surface boundary current is Sea. Deep Sea Res Top Stud Oceanogr 57:441–457 in the opposite sense, anticyclonic around the coastline, Cardin V, Bensi M, Pacciaroni M (2011) Variability of water mass and the deep convection in the basin interior is not longer properties in the last two decades in the South Adriatic Sea with emphasis on the period 2006–2009. Cont Shelf Res supported with reversal of the meridional overturning 31(9):951–965 circulation.

1 3 River runoff influences on the Central Mediterranean overturning circulation

Castellari S, Pinardi N, Leaman K (1998) A model study of Jackett DR, McDougall TJ (1995) Minimal adjustment of hydro- air–sea interactions in the Mediterranean Sea. J Mar Syst graphic data to achieve static stability. J Atmos Ocean Tech 18(1–3):89–114 12:381–389 Cessi P, Pinardi N, Lyubartsev V (2014) Energetics of semi- Janeković I, Mihanović H, Vilibić I, Tudor M (2014) Extreme cooling enclosed basins with two-layer flows at the strait. J Phys and dense water formation estimates in open and coastal regions Oceanogr 44(3):967–979 of the Adriatic Sea during the winter of 2012. J Geophys Res Curchitser EN, Haidvogel DB, Iskandarani M (2001) Transient Oceans 119(5):3200–3218 adjustment of circulation in a midlatitude Abyssal Ocean Jerlov N (1976) Marine Optics. Elsevier Science Publishing Company basin with realistic geometry and bathymetry. J Phys Ocean- Kassis D, Konstantinidou A, Perivoliotis L, Korres G (2016) Inter- ogr 31:725–745 comparing numerical model simulations in the Ionian Sea with Cushman-Roisin B, Gacić M, Poulain P-M, Artegiani A, 2002. Argo T/S profiles for the period 2008–2012 Physical oceanography of the Adriatic Sea: past, present and Killworth PD (2006) Time interpolation of forcing fields in ocean future. Kluwer Academic Publishers, pp 304 models. J Phys Oceanogr 26:136–143 De Lavergne C, Madec G, Le Sommer J, Nurser AG, Naveira Gara- Kondo J (1975) Air–sea bulk transfer coefficients in diabatic condi- bato AC (2016) The impact of a variable mixing efficiency on tions. Bound-Layer Meteor 9:91–112 the abyssal overturning. J Phys Oceanogr 46(2):663–681 Knudsen M (1900) Ein hydrographischer Lehrsatz. Ann Hydrogr Dell’Aquila A, Calmanti S, Ruti P, Struglia MV, Pisacane G, Carillo Maritimen Meteor 28:316–320 A, Sannino G (2012) Effects of seasonal cycle fluctuations in Kourafalou VH, Oey L-Y, Wang JD, Lee TL (1996) The fate of an A1B scenario over the Euro-Mediterranean region. Clim river discharge on the continental shelf. 1. Modeling the river Res 2:135 plume and the inner shelf coastal current. J Geophys Res Federico I, Pinardi N, Coppini G, Oddo P, Lecci R, Mossa M 101:3415–3434 (2016) Coastal ocean forecasting with an unstructured-grid Lascaratos A (1993) Estimation of deep and intermediate water mass model in the Southern Adriatic Northern Ionian Sea. Nat formation rates in the Mediterranean Sea. Deep Sea Res Part II Hazards Earth Syst Sci Discuss. doi:10.5194/nhess-2016-169 Topical Stud Oceanogr 40(6):1327–1332 Ferrari R, Mashayek A, McDougall TJ, Nikurashin M, Campin JM Ludwig W, Dumont E, Meybeck M, Heussner S (2009) River dis- (2016) Turning ocean mixing upside down. J Phys Oceanogr charges of water and nutrients to the Mediterranean and Black Flather RA (1976) A tidal model of the northwest European conti- Sea: major drivers for ecosystem changes during past and future nental shelf. Memories de la Societe Royale des Sciences de decades. Prog Oceanogr 80:199–217 Liege 6(10):141–164 Madec G (2008) NEMO ocean engine. Note du Pole de modélisa- Gacic ́ M, Eusebi Borzelli GL, Civitarese G, Cardin V, Yari S tion, vol 27. Institut Pierre-Simon Laplace (IPSL), France, (2010) Can internal processes sustain reversals of the ocean pp 1288–1619 upper circulation?, The Ionian Sea example. Geophys Res Maggiore A, Zavatarelli M, Angelucci MG, Pinardi N (1998) Surface Lett 37:L09608 heat and water fluxes in the Adriatic Sea: seasonal and interan- Gacic ́ M, Civitarese G, Borzelli, G.L.E., Kovacevic V, Poulain nual variability. Phys Chem Earth 23(5–6):561–567 PM, Theocharis A, Menna M, Catucci A, Zarokanellos N Malacic V, Petelin B (2009) Climatic circulation in the Gulf of Trieste (2011) On the relationship between the decadal oscillations (northern Adriatic). J Geophys Res 114(C7) of the northern Ionian Sea and the salinity distributions in the Malanotte-Rizzoli P, Manca BB, Ribera d’Alcala’ M, Theocharis A eastern Mediterranean. J Geophys Res 116–125 (1997) A synthesis of the Ionian Sea hydrography, circulation Gertman I, Pinardi N, Popov Y, Hecht A (2006) Aegean Sea and water mass pathways during POEM-Phase I. Prog Ocean water masses during the early stages of the Eastern Medi- 39:153–204 terranean climatic transient (1988–90). J Phys Oceanogr Manca B, Kovaĉević V, Gaĉić M, Viezzoli D (2002) Dense water for- 36(9):1841–1859 mation in the Southern Adriatic Sea and spreading into the Ion- Grilli F, Pinardi N (1998) The computation of Rossby radii of ian Sea in the period 1997–1999. J Mar Syst 33–34:133–154 deformation for the Mediterranean Sea. MTP News Manca B, Budillon G, Scarazzato P, Orsella L (2003) Evolution of Guarnieri A, Pinardi N, Oddo P, Bortoluzzi G, Ravaioli M (2013) dynamics in the Eastern Mediterranean affecting water mass Impact of tides in a baroclinic circulation model of the Adri- structures and properties in the Ionian and Adriatic Seas. J Geo- atic Sea. J Geophys Res Oceans 118(1):166–183 phys Res 108(C9):101029–101046 Gunduz M, Dobricic S, Oddo P, Pinardi N (2013) Impact of Levan- Mantziafou A, Lascaratos A (2004) An eddy resolving numerical tine Intermediate Water on the interannual variability of study of the general circulation and deep-water formation in the the Adriatic Sea based on simulations with a fine resolution Adriatic Sea. Deep Sea Res Part I Oceanogr Res Pap Hist Med ocean model. Ocean Model 72:253–263 Assoc 51(7):921–952 Haiden T, Janousek M, Bauer P, Bidlot J, Dahoui M, Ferranti L, Mantziafou A, Lascaratos A (2008) Deep-water formation in the Hewson T, Prates F, Richardson DS, Vitart F (2015) Evalu- Adriatic Sea: Interannual simulation for years 1979–1999. ation of ECMWF forecasts, including 2014–2015 upgrades. Deep Sea Res Part I Oceanogr Res Pap Hist Med Assoc Technical report 765. ECMWF 55(11):1403–1427 Hainbucher D, Rubino A, Klein B (2006) Water mass characteris- Marchesiello P, Williams JM, Shchepetkin A (2001) Open boundary tics in the deep layers of the western Ionian Basin observed conditions for long-term integrations of regional oceanic mod- during May 2003. Geophys Res Lett 33:4 els. Ocean Model 3(1–2):1–20 Hellerman S, Rosenstein M (1983) Normal monthly wind stress Marshall J, Speer K (2012) Closure of the meridional overturning over the world ocean with error estimates. J Phys Oceanogr circulation through Southern Ocean upwelling. Nat Geosci 13:1093–1104 5(3):171–180 Huang RX (1993) Real freshwater flux as a natural boundary con- Masina S, Pinardi N (1994) Mesoscale data assimilation studies in the dition for the salinity balance and thermohaline circulation middle Adriatic Sea. Cont Shelf Res 14(12):1293–1310 forced by evaporation and precipitation. J Phys Oceanogr Medhycos (2001) The Mediterranean hydrological cycle observing 23(11):2428–2446 system. Medhycos phase II, period 2002–2005, report no. 17, pp 36

1 3 G. Verri et al.

Mellor G, Blumberg A (2004) Wave breaking and ocean surface layer Provini A, Crosa G, Marchetti R (1992) Nutrient export from Po thermal response. J Phys Oceanogr 34(3):693–698 and Adige river basins over the last 20 years. Sci Total Envi- Mihanovic H, Vilibic I, Carniel S, Tudor M, Russo A, Bergamasco A, ron Suppl:291–313 Bubic N, Ljubesic Z, Vilicic D, Boldrin A, Malacic V (2013) Rahmstorf S (1995) Bifurcations of the Atlantic thermohaline cir- Exceptional dense water formation on the Adriatic shelf in the culation in response to changes in the hydrological cycle. winter of 2012. Ocean Sci 9:561–572 Nature 378:145–149 Milliman JD (2001) Delivery and fate of fluvial water and sediment Rahmstorf S (1996) On the freshwater forcing and transport of the to the sea: a marine geologist’s view of European rivers. Sci Atlantic thermohaline circulation. Clim Dyn 12(12):799–811 Marina 65(Suppl 2):121–132 Raicich F (1996) Note on the flow rates of the Adriatic rivers. Tech. Munk W, Wunsch C (1998) Abyssal recipes II. Energetics of tidal and Rep. RF 02/94. CNR, Istituto Talassografico di Trieste, Italy, wind mixing. Deep Sea Res Part I Oceanogr Res Pap Hist Med p 8 Assoc 45(12):1977–2010 Robinson AR, Leslie WG, Theocharis A, Lascaratos A (2001) Med- Nikurashin M, Vallis G (2011) A theory of deep stratification iterranean sea circulation. Ocean Curr, 1689–1705 and overturning circulation in the ocean. J Phys Oceanogr Roether W, Manca BB, Klein B, Bregant D, Georgopoulos D, Beit- 41(3):485–502 zel V, Kovacevic V, Luchetta A (1996) Recent changes in Oddo P, Guarnieri A (2011) A Study of the hydrographic conditions eastern Mediterranean deep waters. Science 271:333–335 in the Adriatic Sea from numerical modelling and direct obser- Roussenov VM, Williams RG, Roether W (2001) Comparing the vations (2000–2008). Ocean Sci 7:549–567 overflow of dense water in isopycnic and cartesian mod- Oddo P, Pinardi N, Zavatarelli M (2005) A numerical study of the els with tracer observations in the eastern Mediterranean. interannual variability of the Adriatic Sea (2000–2002). Sci Deep Sea Res Part I Oceanogr Res Pap Hist Med Assoc Total Environ 353(1):39–56 48(5):1255–1277 Oddo P, Adani M, Pinardi N, Fratianni C, Tonani M, Pettenuzzo D Rubino A, Hainbucher D (2007) A large abrupt change in the abyssal (2009) A nested Atlantic-Mediterranean Sea general circulation water masses of the eastern Mediterranean. Geophys Res Lett. model for operational forecasting. Ocean Sci doi:10.1029/2007GL031737 Paparella F, Young WR (2002) Horizontal convection is non-turbu- Rubino A, Romanenkov D, Zanchettin D, Cardin V, Hainbucher D, lent. J Fluid Mech 466:214 Bensi M, Boldrin A, Langone L, Miserocchi S, Turchetto M Pasaric M (2004) Annual cycle of river discharge along the Adri- (2012) On the descent of dense water on a complex canyon sys- atic coast of Croatia. Rapports et procès-verbaux des reunions tem in the southern Adriatic basin. Cont Shelf Res 44:20–29 CIESMM 37:132 Sanchez-Gomez E, Somot S, Josey SA, Dubois C, Elguindi N, Déqué Paschini E, Artegiani A, Pinardi N (1993) The mesoscale eddy field M (2011) Evaluation of the Mediterranean Sea water and heat of the middle Adriatic Sea during fall 1988. Deep Sea Res Part I budgets as simulated by an ensemble of high resolution regional Oceanogr Res Pap Hist Med Assoc 40(7):1365–1377 climate models. Clim Dyn 37:2067–2086 Payne RE (1972) Albedo of the sea surface. J Atmos Sci 29:959–970 Schroeder K, Garcìa-Lafuente J, Josey S, Artale V, Buongiorno- Pedlosky J (1987) Geophysical fluid dynamics. Springer, Berlin, pp Nardelli B, Gacic M, Gasparini GP, Herrmann M, Lionello P, 724 Ludwig W, Millot C, Ozsoy E, Pisacane G, Sanchez-Garrido Pettenuzzo D, Large W, Pinardi N (2010) On the corrections of ERA- JC, Sannino G, Santoleri R, Somot S, Struglia M, Stanev E, 40 surface flux products consistent with the Mediterranean heat Taupier-Letage I, Tsimplis M, Vargas-Yanez M, Zervakis V, and water budgets and the connection between basin surface Zodiatis G (2012) Circulation of the Mediterranean Sea and its total heat flux and NAO. J Geophys Res 115:C06022 variability (Chap. 4). In: Lionello P (ed) Mediterranean climate Pickard GL, Emery WJ (1982) Descriptive physical oceanography: an variability. Elsevier, Amsterdam, pp 187–256 introduction, 4th edn. Pergamon Press, Oxford, pp 265 Simoncelli S, Pinardi N, Oddo P, Mariano AJ, Montanari G, Rinaldi Pinardi N, Coppini G (2010) Operational oceanography in the Medi- A, Deserti M (2011) Coastal rapid environmental assess- terranean Sea: the second stage of development. Ocean Sci ment in the Northern Adriatic Sea. Dyn Atmos Oceans 6:263–267 52(1–2):250–283 Pinardi N, Masetti E (2000) Variability of the large scale general Skliris N, Sofianos S, Lascaratos A (2007) Hydrological changes in circulation of the Mediterranean Sea from observations and the Mediterranean Sea in relation to changes in the freshwater modelling: a review. Palaeogeogr Palaeoclimatol Palaeoecol budget: a numerical modelling study. J Mar Syst 65:400–416 158(3):153–173 Spall MA (2004) Boundary currents and water mass transformation in Pinardi N, Allen I, Demirov E, De Mey P, Korres G, Lascaratos A, marginal seas. J Phys Oceanogr 34:1197–1213 Le Traon PY, Maillard C, Manzella G, Tziavos C (2003) The Spall MA (2010) Dynamics of downwelling in an eddy-resolving con- Mediterranean ocean forecasting system: first phase of imple- vective basin. J Phys Oceanogr 40:2341–2347 mentation (1998–2001). Ann Geophys 21(1):3–20 Spall MA (2011) On the role of eddies and surface forcing in the heat Pinardi N, Arneri E, Crise A, Ravaioli M, Zavatarelli M (2006) The transport and overturning circulation in marginal sea. J Mar Res physical, sedimentary and ecological structure and variability of 24:4844–4858 shelf areas in the Mediterranean Sea. Sea 14:1243–1330 Spall MA (2012) Influences of precipitation on water mass trans- Pinardi N, Zavatarelli M, Adani M, Coppini G, Fratianni C, Oddo P, formation and deep convection. J Phys Ocean 42:1684–1700. Tonani M, Lyubartsev V, Dobricic S, Bonaduce A (2015) Medi- doi:10.1175/jpo-d-11-0230.1 terranean Sea large-scale low frequency ocean variability and Struglia MV, Mariotti A, Filograsso A (2004) River discharge into water mass formation rates from 1987 to 2007: a retrospective the Mediterranean Sea: climatology and aspects of the observed analysis. Prog Oceanogr 132:318–332 variability. J Clim 17:4740–4751 Pisacane G, Artale V, Calmanti S, Rupolo V (2006) Decadal oscilla- Sverdrup HU (1947) Wind-driven currents in a baroclinic ocean; with tions in the Mediterranean Sea: a rFeesult of the overturning cir- application to the equatorial currents of the eastern Pacific. Proc culation variability in the eastern basin? Clim Res 31:257–271 Natl Acad Sci 33:318–326 Pisano A, Nardelli BB, Tronconi C, Santoleri R (2016) The new Med- Tonani M, Pinardi N, Dobricic S, Pujol I, Fratianni C (2008) A high- iterranean optimally interpolated pathfinder AVHRR SST data- resolution free surface model of the Mediterranean Sea. Ocean set (1982–2012). Remote Sens Environ 176:107–116 Sci 4(1):1–14

1 3 River runoff influences on the Central Mediterranean overturning circulation

Ursella L, Kovacevic V, Gacic M (2011) Footprints of mesoscale Vilibić I, Supić N (2005) Dense water generation on a shelf: the case eddy passages in the Strait of Otranto (Adriatic Sea). J Geopys of the Adriatic Sea. Ocean Dyn 55(5–6):403–415 Res 116:14 Vörösmarty C, Fekete B, Tucker B (1996) Global river discharge Vervatis VD, Sofianos SS, Skliris N, Somot S, Lascaratos A, Rixen database, RivDis. UN Educ. Sci. And Cult. Organ, Paris. http:// M (2013) Mechanisms controlling the thermohaline circulation www.rivdis.sr.unh.edu pattern variability in the Aegean–Levantine region. A hindcast Wang XH, Oddo P, Pinardi N (2007) On the bottom density plume simulation (1960–2000) with an eddy resolving model. Deep on coastal zone off Gargano (Italy) in the southern Adriatic Sea Sea Res Part I Oceanogr Res Pap Hist Med Assoc 74:82–97 and its interannual variability. J Geophys Res 112 Vilibić I, Orlić M (2001) Least-squares tracer analysis of water Wu P, Haines K (1996) Modeling the dispersal of levantine interme- masses in the South Adriatic (1967–1990). Deep Sea Res diate water and its role in Mediterranean deep water formation. I(48):2297–2330 J Geophys Res 101:6591–6607 Vilibić I, Orlić M (2002) Adriatic water masses, their rates of formation and transport through the Otranto Strait. Deep Sea Res Part I Oceanogr Res Pap Hist Med Assoc 49(8):1321–1340

1 3